A method, system, and equipment for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization.

Abstract

This invention discloses a method, system, and device for estimating the diameter at breast height (DBH) of individual trees based on multi-stage feature optimization. The method includes: acquiring raw point cloud data from an airborne lidar system over a target study area; obtaining normalized forest point cloud data based on the raw point cloud data, and segmenting the point cloud data of individual trees based on the normalized forest point cloud data to obtain single-tree point cloud data corresponding to multiple individual trees in the target study area; extracting candidate feature sets from each single-tree point cloud data, constructing a candidate feature matrix, and performing linear correlation screening, nonlinear information screening, and stability optimization screening based on the candidate feature matrix to obtain the optimal feature subset; and outputting the corresponding DBH prediction result using a pre-constructed single-tree DBH estimation model based on the feature data of the single tree to be predicted corresponding to the optimal feature subset. This method can progressively eliminate redundant and invalid features while retaining key information, improving the effective utilization of information between features and DBH, and overall enhancing the accuracy of single-tree DBH estimation.

Description

Technical Field

[0001] This invention relates to the field of forest resource remote sensing monitoring technology, and in particular to a method, system and equipment for estimating the diameter at breast height of a single tree based on multi-stage feature optimization. Background Technology

[0002] Diameter at breast height (DBH) is an important structural parameter reflecting the growth status of trees, forest stock volume, and ecological function, and it plays a significant role in forest resource surveys, biomass estimation, and carbon cycle research. Traditional methods of obtaining DBH mainly rely on manual plot measurements, which suffer from high labor intensity, low efficiency, and severe terrain limitations, making it difficult to meet the needs of rapid monitoring of large-scale forest resources. With the development of airborne lidar (LiDAR) technology, LiDAR, with its advantages of high spatial resolution, strong vegetation penetration capability, and rapid acquisition of three-dimensional forest structure information, provides a new technical approach for the inversion of forest structure parameters.

[0003] Although methods for estimating breast diameter based on airborne lidar point clouds have been applied, the high dimensionality and redundant information of point cloud features make it difficult to fully extract effective information related to breast diameter, resulting in insufficient accuracy in breast diameter estimation results. Summary of the Invention

[0004] This invention provides a method, system, and device for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization, in order to solve the defect of "insufficient accuracy in DBH estimation results" in the prior art.

[0005] This invention provides a method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization, comprising: Acquire raw point cloud data from airborne lidar in the target research area; Based on the original point cloud data, normalized forest point cloud data is obtained, and based on the normalized forest point cloud data, single tree point cloud segmentation is performed to obtain single tree point cloud data corresponding to multiple single trees in the target study area. Extract candidate feature sets from each of the single tree point cloud data, construct a candidate feature matrix, and perform linear correlation screening, nonlinear information screening, and stability optimization screening based on the candidate feature matrix to obtain the optimal feature subset; Based on the feature data of the individual tree to be predicted corresponding to the optimal feature subset, the corresponding diameter at breast height (DBH) prediction result is output using the pre-constructed individual tree DBH estimation model.

[0006] According to the present invention, a method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization is provided. The step of obtaining normalized forest point cloud data based on the original point cloud data specifically includes: Separate ground points from non-ground points from the raw point cloud data; Based on the separated ground points, spatial interpolation is performed using the inverse distance weighted interpolation method to obtain complete ground elevation information within the target study area, so as to construct a digital elevation model. The original point cloud data is normalized using the digital elevation model to obtain normalized forest point cloud data.

[0007] According to the present invention, a method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization is provided. The step of segmenting the normalized forest point cloud data into single-tree point clouds to obtain single-tree point cloud data corresponding to multiple single trees in the target study area specifically includes: A canopy height model is constructed based on the normalized forest point cloud data, and the top positions of multiple individual trees are obtained by local extremum detection on the canopy height model. Using the top position of each individual tree as a seed point, individual tree point cloud segmentation is performed to obtain individual tree point cloud data corresponding to multiple individual trees in the target study area.

[0008] According to the present invention, a method for estimating the diameter at breast height of a single tree based on multi-stage feature optimization is provided, wherein the features in the candidate feature set include at least height features, laser echo intensity features, point cloud density features, canopy structure features, and single tree geometric features.

[0009] According to the method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization provided by the present invention, the step of linear correlation screening specifically includes: Calculate the Pearson correlation coefficient between each feature variable in the candidate feature matrix and the diameter at breast height, compare the Pearson correlation coefficient corresponding to each feature variable with a first preset threshold, and remove feature variables whose Pearson correlation coefficient is lower than the first preset threshold.

[0010] According to the method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization provided by the present invention, the nonlinear information filtering step specifically includes: Calculate the mutual information value between each feature variable retained after linear correlation screening and the diameter at breast height. Based on the mutual information value, screen the feature variables and remove feature variables whose mutual information value is lower than a second preset threshold.

[0011] According to the method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization provided by the present invention, the stability optimization and screening step specifically includes: The Boruta algorithm is used to screen the stability of each feature variable retained after nonlinear information filtering in order to determine the optimal feature subset. The Boruta algorithm constructs random perturbation features based on a random forest model, compares each feature variable retained after nonlinear information filtering with the corresponding random perturbation features, and retains feature variables whose importance is higher than the random perturbation features.

[0012] According to the multi-stage feature optimization-based method for estimating the diameter at breast height (DBH) of a single tree provided by the present invention, the step of pre-constructing a DBH estimation model for a single tree specifically includes: The feature data corresponding to the optimal feature subset of each individual tree sample are matched with the measured diameter at breast height (DBH) data to construct a training set; Based on the training set and the XGBoost regression model, a regression relationship model between features and diameter at breast height (DBH) is established to obtain a single-tree DBH estimation model.

[0013] This invention also provides a single-tree diameter-at-breast-width estimation system based on multi-stage feature optimization, comprising: The data acquisition module is used to acquire raw point cloud data from the airborne lidar of the target research area; The point cloud normalization and single tree segmentation module is used to obtain normalized forest point cloud data based on the original point cloud data, and to perform single tree point cloud segmentation based on the normalized forest point cloud data to obtain single tree point cloud data corresponding to multiple single trees in the target study area. The feature extraction and multi-stage feature optimization module is used to extract candidate feature sets of each single tree point cloud data, construct a candidate feature matrix, and perform linear correlation screening, nonlinear information screening and stability optimization screening based on the candidate feature matrix to obtain the optimal feature subset. The diameter at breast height (DBH) estimation module is used to output the corresponding DBH prediction result based on the feature data of the individual tree to be predicted corresponding to the optimal feature subset, using a pre-built single-tree DBH estimation model.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the single-tree diameter-at-breast-length estimation method based on multi-stage feature optimization as described above.

[0015] The present invention provides a method, system, and device for estimating tree diameter at breast height (DBH) based on multi-stage feature optimization. It segments normalized forest point cloud data to obtain point cloud data for multiple individual trees in the target study area, eliminating the influence of terrain undulations on tree height measurement and obtaining independent point cloud data for each tree. Furthermore, it constructs a candidate feature matrix and performs multi-stage screening using linear, non-linear, and stability parameters. This process progressively eliminates redundant and invalid features, avoiding information loss from one-time screening while retaining key information. Finally, based on the feature data of the tree to be predicted corresponding to the optimal feature subset, it outputs the corresponding DBH prediction result using a pre-constructed tree DBH estimation model. This reduces noise interference in high-dimensional features, improves the effective utilization of information between features and DBH, and enhances the accuracy of tree DBH estimation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the method for estimating the diameter at breast height of a single tree based on multi-stage feature optimization provided by the present invention. Figure 2 This is a schematic diagram of the single-tree point cloud segmentation process of the single-tree diameter estimation method based on multi-stage feature optimization provided by the present invention. Figure 3 This is a flowchart of the multi-stage feature optimization process for estimating the diameter at breast height of a single tree based on multi-stage feature optimization provided by the present invention. Figure 4 This is a schematic diagram illustrating the construction process of the single-tree diameter estimation model based on the multi-stage feature optimization method for single-tree diameter estimation provided by the present invention. Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0019] In the description of this invention, it should be noted that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0020] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] See Figure 1 This application provides a method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization, comprising the following steps: S110: Acquire raw point cloud data of the airborne lidar in the target research area; S120: Obtain normalized forest point cloud data based on the original point cloud data, and perform single tree point cloud segmentation based on the normalized forest point cloud data to obtain single tree point cloud data corresponding to multiple single trees in the target study area. S130: Extract candidate feature sets from the point cloud data of each tree, construct a candidate feature matrix, and perform linear correlation screening, nonlinear information screening, and stability optimization screening based on the candidate feature matrix to obtain the optimal feature subset; S140: Based on the feature data of the single tree to be predicted corresponding to the optimal feature subset, the corresponding diameter at breast height (DBH) prediction result is output using the pre-built single tree DBH estimation model.

[0022] This application uses normalized forest point cloud data to segment individual tree point clouds, obtaining individual tree point cloud data for multiple trees in the target study area. This eliminates the influence of terrain undulation on tree height measurement and obtains independent point cloud data for each tree. Furthermore, a candidate feature matrix is ​​constructed and subjected to multi-stage screening using linear, nonlinear, and stability methods. This process progressively eliminates redundant and invalid features, avoiding information loss from a single screening while retaining key information. Finally, based on the feature data of the tree to be predicted corresponding to the optimal feature subset, a pre-constructed individual tree diameter-at-breast-width (DBH) estimation model is used to output the corresponding DBH prediction results. This reduces noise interference in high-dimensional features, improves the effective utilization of information between features and DBH, and enhances the accuracy of individual tree DBH estimation.

[0023] In S110, raw point cloud data of the target study area is acquired by airborne LiDAR. After acquiring the raw point cloud data, outlier removal and noise filtering are performed on the point cloud data to eliminate abnormal points and interference signals.

[0024] For example, a hexacopter drone equipped with a lidar sensor can be used to collect raw point cloud data. The models of the hexacopter drone and lidar sensor are selected according to the actual situation, and this application does not impose specific restrictions.

[0025] In S120, the step of obtaining normalized forest point cloud data based on the original point cloud data specifically includes: separating ground points and non-ground points from the original point cloud data; performing spatial interpolation processing based on the separated ground points using the inverse distance weighted interpolation method to obtain complete ground elevation information within the target study area, so as to construct a digital elevation model; and using the digital elevation model to perform elevation normalization on the original point cloud data to obtain normalized forest point cloud data.

[0026] In this step, a filtering algorithm is used to classify the original point cloud into ground points and non-ground points. For example, the filtering algorithm uses the CSF method. The CSF (Cloth Simulation Filter) method simulates a piece of virtual cloth falling and contacting the point cloud surface. Based on the final shape of the cloth, ground points and non-ground points are identified, thereby separating the terrain and ground features.

[0027] Due to factors such as discrete sampling by lidar and vegetation obstruction, the original ground point cloud is usually discretely distributed and cannot directly form a complete surface model. By interpolation, the ground elevation of unknown points can be estimated, thereby generating a complete digital elevation model (DEM). In this application, the spatial resolution of the DEM is set to 0.5 m × 0.5 m.

[0028] Specifically, spatial interpolation is performed based on the separated ground points using an inverse distance weighted interpolation method: the inverse distance between the point to be interpolated and each known ground point is used as the weight, and the ground elevation of the unknown location is estimated by weighted averaging. Thus, the discrete ground point cloud is converted into a fully covered digital elevation model (DEM). This DEM stores the ground elevation of each location in the form of a regular grid, which is used to characterize the surface topography and provide a topographic benchmark for subsequent point cloud elevation normalization.

[0029] Furthermore, using the constructed digital elevation model, the absolute altitude of each point in the original point cloud is normalized: the elevation value of each point is subtracted from the DEM ground elevation value corresponding to its location, converting the point cloud elevation into its vertical height relative to the ground (i.e., height above the ground). After processing, the ground point height approaches 0, and the vegetation point height reflects the true tree height, thus obtaining normalized forest point cloud data.

[0030] See Figure 2 The steps for segmenting individual tree point clouds based on normalized forest point cloud data to obtain individual tree point cloud data corresponding to multiple individual trees in the target study area include: constructing a canopy height model based on normalized forest point cloud data; obtaining the top positions of multiple individual trees through local extremum detection on the canopy height model; and segmenting individual tree point clouds using the top positions of each individual tree as seed points to obtain individual tree point cloud data corresponding to multiple individual trees in the target study area.

[0031] The Canopy Height Model (CHM) is a raster model describing the vertical height of tree canopies above the ground. It can be calculated from the difference between the Digital Surface Model (DSM) and the Digital Elevation Model (DEM). The DSM is a raster model that includes the top surface height of all objects on the ground and is a rasterized representation of the original point cloud data.

[0032] Gaussian filtering is used to smooth the crown height model (CHM) to reduce the impact of local noise on treetop detection. Based on the crown height model, local maxima are searched within a sliding window of a set size, and these local maxima are used as multiple individual treetop locations.

[0033] Furthermore, the detected treetop locations are used as initial seed points for the PCS algorithm, with each seed point representing a potential tree. Starting from the seed points, the PCS algorithm uses region growing or clustering to progressively group neighboring points into their corresponding tree point cloud clusters based on the spatial distance and height differences between point clouds, ultimately outputting independent tree point cloud data for each tree.

[0034] In this embodiment of the application, in order to avoid interference from understory vegetation such as low shrubs and herbs in the identification of individual trees, only point clouds with a height of 2 meters or more are retained as candidate individual trees, while point clouds with a height of less than 2 meters are filtered out.

[0035] After completing the individual tree segmentation, structural parameters such as tree height, crown diameter, and crown area can be extracted, and the segmentation results can be verified in conjunction with sample plot survey data. The segmentation effectiveness is evaluated using recall (R), precision (P), and F-score (F). Recall is the proportion of correctly segmented trees out of actual trees, precision is the proportion of correctly segmented trees, and the F-score is the harmonic mean of recall and precision, comprehensively reflecting the segmentation effect.

[0036] In step S130, feature extraction is performed on the segmented individual tree point cloud data to construct a candidate feature matrix for diameter at breast height (DBH) estimation. The candidate feature set includes at least height features, laser echo intensity features, point cloud density features, canopy structure features, and individual tree geometric features. Specifically, height features describe the vertical distribution of the individual tree point cloud, point cloud density features describe the spatial distribution of the point cloud, laser echo intensity features describe the variation in laser echo energy, and canopy structure features describe the spatial morphology of the tree canopy. For example, LiDAR360 software can be used to perform statistical analysis on the individual tree point cloud to extract these features.

[0037] In detail, statistical parameters reflecting the vertical structure distribution, spatial density distribution and laser echo response characteristics of trees are extracted from single-tree point cloud data, and candidate feature sets are constructed by combining structural parameters such as single-tree height, crown diameter and crown area.

[0038] Preferably, the candidate feature matrix contains 104-dimensional candidate feature variables. For height, laser echo intensity, point cloud density, and canopy structure features, 101-dimensional point cloud feature variables are extracted. Furthermore, tree height, crown diameter, and canopy area obtained during the individual tree segmentation stage are introduced into the feature set as auxiliary structural parameters. Finally, a candidate feature matrix containing 104-dimensional feature variables is constructed, providing input data for subsequent feature optimization.

[0039] LiDAR point cloud features generally suffer from high dimensionality, strong redundancy, and complex nonlinear coupling relationships. Direct modeling can easily lead to overfitting, poor stability, and decreased estimation accuracy. Existing feature selection methods often employ single-stage correlation analysis or single feature importance ranking, which makes it difficult to simultaneously consider linear correlation, nonlinear coupling, and feature stability, resulting in information gaps and insufficient robustness in the selection results.

[0040] To address the aforementioned issues, based on the candidate feature matrix obtained through the above steps, the feature variables are progressively screened and optimized to obtain the optimal feature subset for estimating the diameter at breast height (DBH) of a single tree. See [link to relevant documentation]. Figure 3The multi-stage feature optimization method adopts the MSFS (Multi-Stage Feature Selection) method, which includes three stages: linear correlation screening, nonlinear information screening, and stability optimization screening.

[0041] The steps of linear correlation screening specifically include: calculating the Pearson correlation coefficient between each feature variable in the candidate feature matrix and the diameter at breast height; comparing the Pearson correlation coefficient corresponding to each feature variable with a first preset threshold; and removing feature variables whose Pearson correlation coefficient is lower than the first preset threshold.

[0042] The formula for calculating the Pearson correlation coefficient is as follows: ; in, Let i be the feature value of the i-th sample. Let be the chest diameter value of the i-th sample. The mean of all sample feature values. This represents the mean of the chest diameter values ​​for all samples.

[0043] Based on the correlation coefficient calculation results, feature variables with correlation coefficients lower than a first preset threshold are removed to reduce linear redundancy in the candidate feature matrix. In this embodiment, the feature dimension is reduced from 104 dimensions to 60 dimensions after filtering.

[0044] The steps of nonlinear information filtering specifically include: calculating the mutual information value between each feature variable retained after linear correlation filtering and the diameter at breast height; filtering the feature variables based on the mutual information value; and removing feature variables whose mutual information value is lower than a second preset threshold.

[0045] Mutual information is calculated for the feature variables after linear correlation screening to obtain the nonlinear correlation information between each feature variable and chest diameter. The mutual information calculation formula is as follows: ; in, This represents the joint probability distribution of the characteristic variable and chest diameter. and Let these represent the marginal probability distributions of the feature variable and the diameter at breast height, respectively. This represents the mutual information value between the characteristic variable and the diameter at breast height.

[0046] Based on the mutual information calculation results, low-information-contribution feature variables are screened, and feature variables that have a non-linear correlation with chest diameter are retained.

[0047] In some examples, the second preset threshold can be a fixed preset threshold. In other examples, the screening threshold is adaptively adjusted based on the distribution characteristics of mutual information values ​​to improve the adaptability of feature screening under different forest types and sample distribution conditions. For example, a dynamic screening threshold can be constructed using the sum of the mutual information mean and standard deviation, which can avoid the problem of erroneous deletion of effective nonlinear features under fixed threshold conditions. In this embodiment, the feature dimension is reduced from 60 dimensions to 30 dimensions after screening.

[0048] The steps of stability optimization screening specifically include: using the Boruta algorithm to screen the stability of each feature variable retained after nonlinear information screening in order to determine the optimal feature subset; wherein, the Boruta algorithm constructs random perturbation features based on the random forest model, compares each feature variable retained after nonlinear information screening with the corresponding random perturbation features, and retains feature variables whose importance is higher than that of the random perturbation features.

[0049] By constructing random perturbation features (random shadow features) and combining them with a random forest model for multiple rounds of iterative comparison, stable contributing features are retained while invalid or unstable features are removed. When the importance of the original feature is higher than that of the random perturbation feature, the corresponding feature variable is retained; when the importance of the original feature is lower than that of the random perturbation feature, the corresponding feature variable is removed. Subsequently, the importance of features is repeatedly evaluated through multiple iterations of training, and the fluctuation of the importance of feature variables is evaluated by combining cross-validation stability analysis, so as to retain stable contributing features.

[0050] It should be noted that the linear correlation screening, nonlinear information screening, and stability optimization screening are all based on multiple individual tree samples with measured diameter at breast height (DBH). After the above processing, a 13-dimensional optimal feature subset is finally obtained.

[0051] See S140. Figure 4 The steps for pre-constructing a single-tree diameter estimation model specifically include: matching the feature data of each single-tree sample corresponding to the optimal feature subset with the measured diameter data to construct a training set; training based on the training set and the XGBoost regression model to establish a regression relationship model between features and diameter to obtain the single-tree diameter estimation model.

[0052] In one example, during the model construction process, a total of 600 individual tree samples were selected to construct the sample dataset, including 500 coniferous forest samples and 100 broadleaf forest samples. The samples were further divided into training and testing sets, with 400 trees used for model training and 200 trees used for model validation.

[0053] An XGBoost regression model was used to iteratively train the training set, establishing a regression relationship between feature variables and breast diameter through gradient boosting. The model's predictive performance was then validated using a test set. The objective function of the XGBoost regression model is as follows: ; in, Represents the loss function. This represents a regularization term for model complexity. Indicates the first The output of the decision tree model corresponding to each iteration. This represents the total number of training samples. Indicates the index of the sample. This represents the true diameter at breast height (DBH) of the i-th sample. This indicates that the i-th sample is in the first t. The predicted value obtained from one round of iterations Let represent the optimal feature subset of the i-th sample.

[0054] Furthermore, by optimizing hyperparameters such as learning rate, maximum tree depth, subsampling ratio, and feature sampling ratio, the model's generalization ability and diameter-to-chest (DBH) estimation accuracy are improved.

[0055] In the actual use of the model, based on the feature data of the individual tree to be predicted corresponding to the optimal feature subset, the pre-built individual tree diameter at breast height (DBH) estimation model is used to output the corresponding DBH prediction results. Furthermore, corresponding DBH estimation data can be generated. For example, the DBH values ​​of each individual tree can be organized into a structured data table or file (such as CSV, database records), containing information such as individual tree number, location, and DBH value, for use in forest resource surveys, stand structure analysis, biomass estimation, and carbon storage assessment.

[0056] To verify the effectiveness of this method, comparative experiments were conducted using the Decision Tree Regression (DTR) model, the Random Forest Regression (RFR) model, and the Gradient Boosting Regression (GBM) model. Experimental results show that the model's predictive performance continuously improves as the feature optimization process progresses. Specifically, after Pearson filtering, the feature dimension was reduced to 60 dimensions; after mutual information filtering, it was reduced to 30 dimensions; and after Boruta filtering, a 13-dimensional optimal feature subset was obtained. The MSFS-XGBoost model (combining Multi-Stage Feature Optimization (MSFS) and XGBoost Regression) constructed under the 13-dimensional feature condition achieved good results in estimating breast diameter, with a test set determination coefficient R² of 0.901 and a root mean square error (RMSE) of 1.647.

[0057] Compared with existing technologies, this invention has the following advantages: Estimating the diameter at breast height (DBH) of individual trees based on raw point cloud data from airborne lidar reduces the reliance on extensive field surveys in traditional manual measurement methods, improving the data acquisition efficiency of forest resource surveys. By extracting multiple types of parameters, such as height features, laser echo intensity features, point cloud density features, and canopy structure features, a candidate feature set for DBH estimation is constructed, which is beneficial for improving the utilization of individual tree structure information. This application proposes a multi-stage feature optimization method, which uses linear correlation screening, nonlinear information screening, and stability optimization screening to progressively screen high-dimensional features, reducing the impact of feature redundancy on the model training process. Simultaneously, the introduction of dynamic threshold adjustment and stability constraint mechanisms during feature screening improves the consistency and stability of feature screening results under different sample conditions. The use of an XGBoost regression model to construct the DBH estimation model enables the establishment of a regression relationship between feature variables and DBH, improving the DBH estimation capability under complex feature conditions. Therefore, the DBH estimation method for individual trees developed in this application can be applied to forest resource surveys, stand structure analysis, and forest parameter estimation, possessing practical engineering application value.

[0058] The following describes the single-tree diameter at breast height (DBH) estimation system based on multi-stage feature optimization provided by this invention. The single-tree DBH estimation system based on multi-stage feature optimization described below can be referred to in correspondence with the single-tree DBH estimation method based on multi-stage feature optimization described above.

[0059] This application provides a single-tree diameter-at-breast-width estimation system based on multi-stage feature optimization, including: The data acquisition module is used to acquire raw point cloud data from the airborne lidar of the target research area; The point cloud normalization and single tree segmentation module is used to obtain normalized forest point cloud data based on the original point cloud data, and to perform single tree point cloud segmentation based on the normalized forest point cloud data to obtain single tree point cloud data corresponding to multiple single trees in the target study area. The feature extraction and multi-stage feature optimization module is used to extract candidate feature sets from each tree point cloud data, construct a candidate feature matrix, and perform linear correlation screening, nonlinear information screening, and stability optimization screening based on the candidate feature matrix to obtain the optimal feature subset. The diameter at breast height (DBH) estimation module is used to output the corresponding DBH prediction result based on the feature data of the individual tree to be predicted corresponding to the optimal feature subset, using a pre-built single-tree DBH estimation model.

[0060] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communication interface 520, a memory 530, and a communication bus 540. The processor 510, communication interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions from the memory 530 to execute a single-tree diameter estimation method based on multi-stage feature optimization.

[0061] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0062] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization, characterized in that, include: Acquire raw point cloud data from airborne lidar in the target research area; Based on the original point cloud data, normalized forest point cloud data is obtained, and based on the normalized forest point cloud data, single tree point cloud segmentation is performed to obtain single tree point cloud data corresponding to multiple single trees in the target study area. Extract candidate feature sets from each of the single tree point cloud data, construct a candidate feature matrix, and perform linear correlation screening, nonlinear information screening, and stability optimization screening based on the candidate feature matrix to obtain the optimal feature subset; Based on the feature data of the individual tree to be predicted corresponding to the optimal feature subset, the corresponding diameter at breast height (DBH) prediction result is output using the pre-constructed individual tree DBH estimation model.

2. The method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization according to claim 1, characterized in that, The step of obtaining normalized forest point cloud data based on the original point cloud data specifically includes: Separate ground points from non-ground points from the raw point cloud data; Based on the separated ground points, spatial interpolation is performed using the inverse distance weighted interpolation method to obtain complete ground elevation information within the target study area, so as to construct a digital elevation model. The original point cloud data is normalized using the digital elevation model to obtain normalized forest point cloud data.

3. The method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization according to claim 1, characterized in that, The step of segmenting individual tree point clouds based on the normalized forest point cloud data to obtain individual tree point cloud data corresponding to multiple individual trees in the target study area specifically includes: A canopy height model is constructed based on the normalized forest point cloud data, and the top positions of multiple individual trees are obtained by local extremum detection on the canopy height model. Using the top position of each individual tree as a seed point, individual tree point cloud segmentation is performed to obtain individual tree point cloud data corresponding to multiple individual trees in the target study area.

4. The method for estimating the diameter at breast height of a single tree based on multi-stage feature optimization according to claim 1, characterized in that, The features in the candidate feature set include at least height features, laser echo intensity features, point cloud density features, canopy structure features, and individual tree geometric features.

5. The method for estimating the diameter at breast height (DBH) of a single tree based on multi-stage feature optimization according to claim 1, characterized in that, The linear correlation screening steps specifically include: Calculate the Pearson correlation coefficient between each feature variable in the candidate feature matrix and the diameter at breast height, compare the Pearson correlation coefficient corresponding to each feature variable with a first preset threshold, and remove feature variables whose Pearson correlation coefficient is lower than the first preset threshold.

6. The method for estimating the diameter at breast height of a single tree based on multi-stage feature optimization according to claim 5, characterized in that, The steps of the nonlinear information filtering specifically include: Calculate the mutual information value between each feature variable retained after linear correlation screening and the diameter at breast height. Based on the mutual information value, screen the feature variables and remove feature variables whose mutual information value is lower than a second preset threshold.

7. The method for estimating the diameter at breast height of a single tree based on multi-stage feature optimization according to claim 6, characterized in that, The stability optimization screening steps specifically include: The Boruta algorithm is used to screen the stability of each feature variable retained after nonlinear information filtering in order to determine the optimal feature subset. The Boruta algorithm constructs random perturbation features based on a random forest model, compares each feature variable retained after nonlinear information filtering with the corresponding random perturbation features, and retains feature variables whose importance is higher than the random perturbation features.

8. The method for estimating the diameter at breast height of a single tree based on multi-stage feature optimization according to claim 1, characterized in that, The steps for pre-constructing a single-tree diameter estimation model specifically include: The feature data corresponding to the optimal feature subset of each individual tree sample are matched with the measured diameter at breast height (DBH) data to construct a training set; Based on the training set and the XGBoost regression model, a regression relationship model between features and diameter at breast height (DBH) is established to obtain a single-tree DBH estimation model.

9. A single-tree diameter-at-breast-width estimation system based on multi-stage feature optimization, characterized in that, include: The data acquisition module is used to acquire raw point cloud data from the airborne lidar of the target research area; The point cloud normalization and single tree segmentation module is used to obtain normalized forest point cloud data based on the original point cloud data, and to perform single tree point cloud segmentation based on the normalized forest point cloud data to obtain single tree point cloud data corresponding to multiple single trees in the target study area. The feature extraction and multi-stage feature optimization module is used to extract candidate feature sets of each single tree point cloud data, construct a candidate feature matrix, and perform linear correlation screening, nonlinear information screening and stability optimization screening based on the candidate feature matrix to obtain the optimal feature subset. The diameter at breast height (DBH) estimation module is used to output the corresponding DBH prediction result based on the feature data of the individual tree to be predicted corresponding to the optimal feature subset, using a pre-built single-tree DBH estimation model.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for estimating the diameter at breast height of a single tree based on multi-stage feature optimization as described in any one of claims 1 to 8.

Images